This is essentially an extended update to my last D-Wave post. Rather than stick it there, I think it is important enough to merit its own post. The reason being, I wish I could make anybody who plans on writing anything on D-Wave first watch the video below from the first Q+ Google+ hang-out this year.
It summarizes the results of the paper I blogged about in my last post on the matter. Ultimately, it answers what is objectively known about D-Wave’s machine based on the analyzed data, and sets out to answer three questions.
- Does the machine work?
- Is is quantum or classical?
- Is it faster than a classical computer?
The short version is
- Based on their modeling D-Wave 2 is indeed a true Quantum Annealer.
- While it can beat an off the shelf solver it cannot (yet) outperform on average a highly targeted hand-crafted classical algorithm.
Of course there is much more in the video, and I highly recommend watching the whole thing. It comes with a good introduction to the subject, but if you only want the part about the test, you may want to skip 11 minutes into the video (this way you also cut out some of the cheap shots at completely clueless popular media reports – an attempt at infusing some humor into the subject that may or may not work for you).
With regards to point (2) the academic discussion is not settled. A paper with heavyweight names on it just came out (h/t Michael Bacon). It proposes a similar annealing behavior could be accomplished with a classical set-up after all. Too me this is truly academic in the best and worst sense i.e. a considerable effort to get all the i’s dotted and the t’s crossed. It simply seems a bit far fetched that the company would set out to build a chip with coupled qubits that behave like a quantum annealer, yet somehow end up with an oddly behaving classical annealer.
From my point of view it is much more interesting to explore all the avenues that are open to D-Wave to improve their chip, such as this new paper on strategies for a quantum annealer to increase the success probability for hard optimization problems. (h/t Geordie Rose).
Geordie Rose weighs in on the paper that claims that the D-Wave machine can be explained classically. He expected a Christmas present and felt he only got socks …
Helmut Katzgraber et al. propose in this paper that the current benchmarks are using the wrong problem set to possibly find a quantum speed-up with D-Wave’s machine.